Soil testing is an important tool for estimating soil K availability and determining how much fertilizer must be applied to realize crop yield potential and minimize fertilizer costs. Our primary objectives were to correlate relative rice (Oryza sativa L.) yield with Mehlich‐3 and 1 mol L−1 HNO3 extractable K, define sufficient whole‐plant K concentrations at panicle differentiation (PD) and early heading (HDG), and calibrate K‐fertilizer rates with Mehlich‐3 soil K in the direct‐seeded, delayed‐flood rice production system. Potassium rate experiments were established at 32 site‐years on silt loams in eastern Arkansas. Relationships between selected parameters were evaluated with linear, curvilinear, and linear‐plateau models. The relationships between relative yield regressed against Mehlich‐3 and HNO3 extractable K were significant and curvilinear. The final curvilinear models for Mehlich‐3 and HNO3 K explained 63 and 43% of yield variability among site‐years, with predicted critical soil concentrations to produce 95% relative yield of 99 and 390 mg K kg−1, respectively. Linear‐plateau models provided comparable critical soil K concentrations. Rice having whole‐plant K concentrations of 23.1 g kg−1 at PD and 13.0 g kg−1 at HDG were predicted to produce 95% relative yield. The predicted K‐fertilizer rates required to optimize rice grain yield depended on the model and ranged from 51 to 90, 41 to 70, 30 to 55, and 20 to 35 kg K ha−1 for soil having Mehlich‐3 soil K concentrations of 60, 70, 80, and 90 mg K kg−1, respectively. The suggested Mehlich‐3 soil and plant K critical concentrations should be appropriate for other U.S. mid‐South rice‐producing areas using similar cultural production practices.